fix(server): honor custom Claude model options #22792 · pingdotgg/t3code

Scope and approval

Fixes #16672

Triaged bug via the automated issue flow.

Evidence index and commands · 10-row behavior matrix · 223-test log. Final head b3f248adfd adds test fixtures only; its production tree exactly matches the after-runtime SHA. Additional SDK harness fixtures cover custom defaults/validation, boolean options, and prompt effort on send/steer.

Verification

Two added behavioural regressions fail on main and pass with the fix: custom Thinking Off reaches SDK settings with effort High preserved; unchanged Thinking reuses a query, and both toggle directions replace it. The implementation step passed 223 focused tests, targeted lint/formatting, and server typecheck. Final independent reviews at the current head returned MERGE from Codex GPT-6.1 Sol and Claude Opus 5.5. Proof audit pass 2 returned SUFFICIENT for all 10 matrix rows and independently reproduced 223 passing tests. The Claude reviewer also checked base production against the new tests: 11 fail on base; two deliberate controls pass.

Evidence

Before	After
Runtime transcript, SHA 74541bab17	Runtime transcript, SHA 95d01097c4

Before

After

Runtime transcript, SHA 74541bab17

Runtime transcript, SHA 95d01097c4

Real browser → isolated dev server → real Claude SDK. Custom dated Sonnet model saves effort High and Thinking Off. Before: query.open has effort high and showThinkingSummaries:true, no alwaysThinkingEnabled. After: false → true → false launches supply the selected boolean and keep effort high; all custom turns complete. The built-in Haiku control supplies alwaysThinkingEnabled:false before the fix; its upstream request returns unknown-provider.

No UI/render paths change. What else changed: custom option descriptors now supply parent launch, query identity, and prompt effort compilation; built-in mappings and existing subagent catalog behavior are retained.

Unverified: macOS, Windows, Electron shell, mobile, relay/tunnel, dynamic upstream catalog refresh; built-in Haiku upstream completion. This uses configured custom model capabilities, not the broader live catalog proposal in #16583.

Model/harness: GPT-6.1 Sol through Codex in T3 Code.

Ggithub-actionsBot8d ago

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	21.6 KiB	21.6 KiB	0 B (0.0%)	24.4 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.3 KiB	1.3 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.9 KiB	21.9 KiB	0 B (0.0%)	24.4 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: da8cc59 · PR result: 8d17b5e · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

Codex decoded thread snapshot: 108.5 KiB
Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

fix(server): honor custom Claude model options#22792

Problem

Change

Scope and approval

Verification

Evidence

Checks

Thread transfer impact